AT&T Performance and Capacity review
Patricio Villar
February 1, 2016
Index
- Performance analysis
- Capacity analysis
- QOS Strategy
- Cost Effectiveness
Performance Analysis:
Non-agregated reports show utilization spikes and sustained utilization during backup windows (small windows on week days and long ones on weekends). Given that congestion control hasn’t been defined, packets are discarded by tail drop in egress queues when contention occurs. Packet discard rate doesn’t exceed the 1% threshold per circuit (~0.01 avg.).
Circuit ID #94/ODGS/200108//OB - Router IEPUDALTX02R1:
Interface Utilization:
Packet Discard Rate:
Circuit ID #94/ODGS/200102//OB - Router IEPUDALTX02R2:
Interface Utilization:
Packet Discard Rate:
Conclusion: (TBD)
Capacity Analysis:
Historically, DCI circuits were unutilized. They main purpose was to allow SNI flows to enter to the Dev/QA site and be ready for any replication flow to take place. With the enablement of DC-to-DC backup/vaulting functionality, VDI cross DC pools and some other miscellaneous cross DC apps, inter-DC traffic has been growing for the last 3 years.
Circuit ID #94/ODGS/200108//OB - Router IEPUDALTX02R1:
Interface Utilization:
Circuit ID #94/ODGS/200102//OB - Router IEPUDALTX02R2:
Interface Utilization:
Regression Analysis
reg_ckt1 <- lm(data = DallasY_ckt1, formula = DallasY_ckt1$Interface...Out.Utilization....~time_index_Y)
reg_ckt2 <- lm(data = DallasY_ckt2, formula = DallasY_ckt2$Interface...Out.Utilization....~time_index_Y)
print(reg_ckt1) ##
## Call:
## lm(formula = DallasY_ckt1$Interface...Out.Utilization.... ~ time_index_Y,
## data = DallasY_ckt1)
##
## Coefficients:
## (Intercept) time_index_Y
## -2.4912 0.7801print(reg_ckt2)##
## Call:
## lm(formula = DallasY_ckt2$Interface...Out.Utilization.... ~ time_index_Y,
## data = DallasY_ckt2)
##
## Coefficients:
## (Intercept) time_index_Y
## 22.1293 0.8863Conclusion:
Analysis shows that the rate of growth per circuit is close to 0.08% (~10% Anual). This makes 2016 a good point in time to analyze different options, as ckt #2 utilization should be at 60-65% utilization in 24 months. 70% Avg. utilization is the recommended point of data circuit upgrade*.
(*) This assumes that the traffic pattern in the DCI will no be affected by any new Application/Service in the next 24 months.
QOS Strategy:
Summary: Currently no QOS profiling is running between Dallas and Columbus DCs. Every application flow between these geographic sites has no priority over another flow. This might not be critical at this point, since link contention only happens sporadically and only near real time apps make users “feel” a bad experience should packet drops occur.
Traffic Accounting reports (Netflow):
Circuit ID #94/ODGS/200108//OB - Router IEPUDALTX02R1:
Top 30 Conversations:
Top 20 Apps:
Top 30 Endpoints:
Circuit ID #94/ODGS/200102//OB - Router IEPUDALTX02R2:
Top 30 Conversations:
Top 20 Apps:
Top 30 Endpoints:
Conclusion: is it highly recommendable to deploy a QOS and backup strategy. At this time, should one link fail backup traffic will take over the remaining link completely, generating potential issues to SNI mgmt and VDI traffic. Backup windows should be reviewed and utilization limits established to avoid link overutilization or excessive oversubscription.
Top talkers:
::Dallas/Columbus Data Domains
::Dallas/Columbus BPNS Netapp filers
Cost Efectiveness:
Summary: Based on current and predicted link utilization patters it is possible to find a cost model using optimization techniques that minimizes operational expenditures. This section should be viewed as a business exercise, pricing information should be used as reference only.
Optimization Analysis Plot:
Conclusion: Do analyze alternatives to “Flat” billing when resizing the current circuits. As demostrated; if historical utilization data is available, better alternatives can be found. When using variable billing models traffic accounting should be part of the operational policy as it directly impacts on monthly fees.